Several research groups are implementing analog integrated circuit models of biological auditory processing. The outputs of these circuit models have taken several forms, including video format for monitor display [1,2], simple scanned output for oscilloscope display [3], and parallel analog outputs suitable for data-acquisition systems [4]. In this paper, we describe an alternative output method for silicon auditory models, suitable for direct interface to digital computers. As a prototype of this method, we describe an integrated circuit model of temporal adaptation in the auditory nerve, that functions as a peripheral to a workstation running the Unix operating system. We show data from a working hybrid system that includes the auditory model, a digital interface, and asynchronous software; this system produces a real-time X-Windows display of the response of the auditory nerve model. interface method. This chip contains both the analog processing and the digital interface circuits; analog signals are not sent off chip.
Communication in Neural SystemsBiological neurons communicate long distances using a pulse representation. Communications engineers have developed several schemes for communicating on a wire using pulses as atomic units. In these schemes, maximally using the communications bandwidth of a wire implies the mean rate of pulses on the wire is a significant fraction of the maximum pulse rate allowed on the wire.Using this criteria, neural systems use wires very inefficiently. In most parts of the brain, most of the wires are essentially inactive most of the time. If neural systems are not organized to fully utilize the available bandwidth of each wire, what does neural communication optimize? Evidence suggests that energy conservation is an important issue for neural systems. A simple strategy for energy conservation is the reduction of the total number of pulses in the representation. Many possible coding strategies satisfy this energy requirement.The strategies observed in neural systems share another common property. Neural systems often implement a class of computations in a manner that produces an energyefficient output encoding as an additional byproduct. The energy-efficient coding is not performed simply for communication and immediately reversed upon receipt, but is an integral part of the new representation. In this way, energy-efficient neural coding is intrinsically different from engineering data compression techniques. Temporal adaptation, lateral inhibition, and spike correlations are examples of neural processing methods that perform interesting computation while producing an energy-efficient output code.These representational principles are the foundation of the neural computation and communication method we advocate in this paper. In this method, the output units of a chip are spiking neuron circuits that use energy-efficient coding methods. To communicate this code off a chip, we use a distinctly non-biological approach.
The Event-Address ProtocolThe unique characteristics of ener...
